A MACHINE LEARNING GEOBAROMETER FOR MAJORITIC GARNET INCLUSIONS IN DIAMOND (Invited Presentation)

Inclusions of majoritic garnet hosted in sub-lithospheric diamonds provide unique insights into processes occurring in Earth’s deep upper mantle and transition zone. Unlike other types of mineral inclusions found in sub-lithospheric diamonds, majoritic garnets can potentially provide accurate estimates of their formation depths because garnet chemistry is known from laboratory experiments to be a strong function of pressure [e.g. 1,2]. However, on the basis of their application to a large compilation of experimental data we find that the available empirical majorite garnet barometers can have large uncertainties, up to 10 GPa in some cases, and so cannot be applied with confidence to diamond-hosted inclusions. To address the challenge of calibrating to all the experimental data we have developed a novel type of majorite barometer using machine learning (ML) algorithms. We applied multiple cross validation strategies and demonstrate that ML fitting using Random Forest Regression significantly outperforms all available barometers, and allows accurate prediction of the formation pressure across the full range of experimental majoritic garnet compositions found in the literature. When our new ML barometer is applied to the global database of diamond-hosted majoritic garnet inclusions we find that their formation occurs in specific pressure modes throughout the upper mantle, with the most prominent peak at about 15 GPa (~450 km). However, exsolved clinopyroxene is often observed within majoritic garnet inclusions and must be accounted for to accurately determine formation pressure. Reconstruction of inclusions in the cases where this is currently possible reveals that ignoring small exsolved components can lead to underestimating inclusion pressures by up to 8 GPa (~ 240 km). Corrected formation pressures of majoritic garnet inclusions from South America place nearly all inclusions formation pressures between 14 and 23 GPa, consistent with models of their crystallization from carbon-rich slab-derived melts at transition zone depths. [1] Akaogi, M., and S. Akimoto (1977), PEPI, 15, 90–106;[2] Beyer, C., and D. J. Frost (2017), EPSL, 461, 30–39.